When in peril, retrench: Testing the portfolio channel of contagion · 2018-10-01 · When in...
Transcript of When in peril, retrench: Testing the portfolio channel of contagion · 2018-10-01 · When in...
-
Journal of International Economics 69 (2006) 203–230
www.elsevier.com/locate/econbase
When in peril, retrench: Testing the portfolio channel
of contagionB
Fernando A. Broner a,e,*, R. Gaston Gelos b, Carmen M. Reinhart c,d
a CREI and Universitat Pompeu Fabra, Ramon Trias Fargas, 25-27, 08005 Barcelona, Spainb International Monetary Fund, 700 19th Street, N.W., Washington, DC 20431, United Statesc Department of Economics, University of Maryland, College Park, MD 20742, United States
d NBER, USAe CEPR, United Kingdom
Received 27 August 2004; received in revised form 18 April 2005; accepted 17 May 2005
Abstract
The impact of past gains and losses on international investors’ risk aversion is an important
factor in the propagation of financial shocks across countries. We first present a stylized model
illustrating how changes in investors’ risk aversion affect portfolio decisions and stock prices.
We then examine empirically the behavior of international mutual funds. When funds’ returns
are below average, they reduce their exposure to countries in which they were overweight and
vice versa. An index of bfinancial interdependenceQ that reflects the extent to which countries
0022-1996/$ -
doi:10.1016/j.
B The views
IMF or IMF p
Olivier Gourin
Juan Solé, Je
Diego), the Fo
Markets Conf
Economics.
* Correspon
Spain. Tel.: +3
E-mail add
see front matter D 2005 Elsevier B.V. All rights reserved.
jinteco.2005.05.004
expressed in this paper are those of the authors and do not necessarily represent those of the
olicy. The authors are grateful to Guillermo Vuletin for able research assistance and to Pierre–
chas, Alejandro Justiniano, Laura Kodres, Miguel Messmacher, Paolo Pesenti, Raghuram Rajan,
remy Stein, and seminar participants at the Bank of Spain, the AEA Annual Meetings (San
urth Annual IMF Research Conference, and the FRBSF and University of Maryland Emerging
erence for helpful comments. Broner acknowledges financial support from CREA-Barcelona
ding author. CREI, Universitat Pompeu Fabra, Ramon Trias Fargas, 25-27, 08005 Barcelona,
4 93 542 2601; fax: +34 93 542 1860.
ress: [email protected] (F.A. Broner).
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230204
share overexposed funds helps explain the pattern of stock market comovement across countries
and the pattern of contagion during crises.
D 2005 Elsevier B.V. All rights reserved.
Keywords: Contagion; International investors; Risk aversion; Emerging markets; Portfolio choice; Financial crises
JEL classification: F30; G15
Through what channels do financial crises spread across countries? Largely owing to
the numerous currency and banking crises of the past decade, a growing body of research
has been devoted to answering this question. While some papers have pointed to trade
linkages as the channel through which contagion is transmitted, there is a growing
consensus that financial linkages and frictions are likely to play a more central role in the
propagation of shocks across national borders.1
At the theoretical level, various authors have sought to explain international financial
contagion with models of investor portfolio choice. Schinasi and Smith (2000) highlight
that contagion effects can be the result of simple portfolio rebalancing within a mean-
variance or VaR framework. In Kodres and Pritsker (2002), differentially informed
investors transmit idiosyncratic shocks from one market to others by rebalancing their
portfolios’ exposures to common macroeconomic risks. Kyle and Xiong (2001) model
contagion as a wealth effect in a model with two risky assets and different types of traders.
Wealth effects as a source of contagion also figure prominently in the models of Goldstein
and Pauzner (2001) and Yuan (2004). In a different approach, Calvo and Mendoza (2000)
describe fund managers’ investment decisions using a mean-variance framework with
short-selling constraints, including fixed costs of information acquisition about countries
and assuming that fund managers’ performance schemes create incentives against
deviating too much from benchmark indices.2
Empirically, there are also some indications that financial links matter. Kaminsky and
Reinhart (2000), Hernández and Valdés (2001), Van Rijckeghem and Weder (2001), and
Caramazza et al. (2000) provide evidence that countries that borrow from the same
creditor as the initial crisis country are more vulnerable to contagion than those
countries that borrow from other creditors that are not engaged in lending to the crisis
country. Providing empirical support for Calvo and Mendoza’s model, Disyatat and
Gelos (2001) show that emerging market funds’ asset allocation can be well
approximated by model with short-sale constraints and mean-variance optimization
around benchmark indices. Van Rijckeghem and Weder (2003) provide evidence that
bank exposures to crisis countries can help predict flows to third countries after the
Mexican and Asian crises.
1 See Kaminsky et al. (2003) for a recent discussion of the evidence on contagion.2 Allen and Gale (2000) analyze how liquidity shocks can be transmitted across banks, leading to contagion of
banking crises. Bernardo and Welch (2004) show that such crises are also possible in asset markets since, in the
absence of perfect execution sequentiality, investors cannot be sure of whether they are selling before or after
prices collapse.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 205
These papers, however, do not explicitly identify the particular mechanism(s) that
accounts for this phenomenon and none of these studies has used cross-sectional
information of portfolio positions at the micro level to identify more precisely the nature of
financial linkages. For example, the studies stressing common lender effects through
commercial banks are based on aggregate information on bank positions, as reported by
the Bank of International Settlements (BIS).
In this paper, we study the trading behavior of emerging market mutual funds and the
role it plays in the transmission of shocks across countries. We take advantage of a large
database of emerging market funds that contains disaggregated information on the
investments of hundreds of funds. For each fund, the database contains monthly data on its
asset allocation by country for the period January 1996 through December 2000. This
detailed information allows us to characterize the behavior of international investors to a
greater extent than was possible in previous studies. Our analysis takes advantage of the
heterogeneity in portfolios and investment behavior across mutual funds. This
heterogeneity implies not only that funds are affected to different extents by country
shocks, but also that the resulting portfolio reallocations transmit these shocks to some
countries more than to others.3
The paper proceeds as follows: In Section 1, we present a stylized model that facilitates
the interpretation of our empirical results. The model incorporates three main ingredients:
(i) investors have heterogeneous beliefs and, thus, hold heterogeneous portfolios; (ii)
investors care about their performance relative to that of other investors; and (iii)
portfolio decisions affect stock prices. In this setting, we analyze the effect of changes in
investors’ risk aversion on portfolio decisions and stock prices. If we further assume that
risk aversion is a function of past relative performance, the model suggests that: (i) in
response to weak past relative performance an investor would shift his/her portfolio
towards the average portfolio, selling assets of countries in which he/she is
boverexposedQ, and buying the assets of countries in which the portfolio isbunderexposedQ; and (ii) crises are transmitted through common overexposed investors,as these effects have their greatest impact on those investors who are most exposed to the
crisis country.4
In the empirical analysis that takes up the remainder of the paper (after describing the
data in Section 2) we examine the effect of gains and losses on investors’ behavior as
regards their portfolio choices. We construct a time-varying index of bfinancialinterdependenceQ, based on the extent to which countries share overexposed funds andassess whether it helps in explaining the transmission of shocks. We do this both
continuously to assess the transmission mechanism of shocks in general, and specifically
during the Thai (1997), Russian (1998), and Brazilian (1999) crises, to examine who
suffered most from contagion effects.
3 Mutual funds represent a non-trivial fraction of equity flows to emerging markets. For example, the funds in
our sample are responsible for 10% of total portfolio equity flows to developing countries in 1998. In addition,
their behavior is likely representative of the behavior of other types of investors as well.4 In the model, investors care about both absolute returns and returns in excess of those of other investors. The
model is related to, but simpler than, models in which investors’ utility is a decreasing function of the variance of
their excess returns over that of other investors (tracking error variance). See Disyatat and Gelos (2001).
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230206
Our main findings can be summarized as follows:
First, we find that when the returns of a particular fund are low relative to the returns of
other funds, the underperforming fund’s reaction is to reduce its weight in countries in
which it was overexposed and increase its weight in countries in which it was
underexposed, thereby adjusting its portfolio in the direction of the average portfolio.
We interpret this result as suggesting that past performance has an effect on funds’ risk
aversion, and that changes in risk aversion affect fund portfolios in the direction suggested
by the model.5,6
Second, the evidence suggests that our financial interdependence index helps explain
stock market comovement across emerging markets above and beyond trade linkages. The
index aids in explaining the transmission of stock market shocks throughout the sample
period. Moreover, there is a negative correlation between countries’ stock market
performance during crises and the degree to which these countries shared overexposed
funds with the initial crisis country. The effect of the financial interdependence index
remains significant in a variety of specifications even after controlling for trade or
commercial bank linkages.
Lastly, we conclude that the predictive power of our index of financial exposure based
on international mutual funds likely reflects the fact that these funds are representative of
other kinds of investors, such as commercial and investment banks.
Taken together, our findings may help explain why some countries are affected by
financial crises in other countries and some are not. Indeed, the financial transmission
mechanism stressed in this paper is present even when countries do not have weak
fundamentals or common features with the initial crisis country. These results suggest that
policymakers might benefit from closely monitoring the micro-composition of investments
across funds in order to become aware which countries may be most vulnerable to
contagion.
These findings may also have interesting implications for understanding momentum
trading at the country level. The fact that, in response to below-average overall
performance, funds tend to reduce their investments in countries in which they are
overexposed can account for the observation that, in the aggregate, funds reduce their
investments in countries in which returns are low (positive-feedback trading).7 The reason
is that when returns in a country are low, funds that are overexposed to that country tend to
have below-average gains. As a result, they reduce their exposure to all countries in which
they are overexposed, including the affected country. Likewise, the funds whose gains are
5 Such changes in risk aversion may result from a wealth effect or be due to compensation schemes for
managers that strongly penalize losses in excess of the industry average, such as hypothesized in Calvo and
Mendoza (2000). There is a substantial literature examining the risk-taking behavior of domestic U.S. fund
managers in response to prior performance (see Chevalier and Ellison, 1997; Brown et al., 1996; Daniel and
Wermers, 2000, among many others). Although this is not the focus of our paper, a discussion of these issues is
provided in Appendix A. More generally, changes in risk aversion by investors have occasionally been cited as a
possible source of contagion. See, for example, Kumar and Persaud (2001).6 Broner et al. (2003) show that the behavior of the term structure of emerging market sovereign bonds also
suggests that investors’ risk aversion increases during crises.7 Among others, Borensztein and Gelos (2003), Kaminsky et al. (2004), and Froot et al. (2001) present
evidence of positive feedback trading in emerging markets.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 207
above average further reduce their exposure to countries in which they are underexposed,
including the affected country. Both effects lead to positive feedback trading in the
aggregate.
1. A stylized model
In this section, we present a stylized model to help in the interpretation of our empirical
results on fund behavior and the transmission of crises. The model has three main features:
investors have heterogeneous beliefs and, thus, hold heterogeneous portfolios; they care
about their performance relative to that of other investors; and portfolio decisions affect
stock prices. The model explores the effect of changes in an investor’s risk aversion on his
portfolio decisions and stock prices.
We assume that investors hold different portfolios because they have different beliefs
about expected dividends. Investors agree to disagree, in the sense that they choose to
ignore the beliefs of other investors although these may be reflected in prices.8 We also
assume that investors are risk averse and may differ in their levels of risk aversion.9 The
existence of heterogeneity across these two dimensions, beliefs and risk aversion, are
necessary to show how a change in an investor’s risk aversion affects his portfolio
decisions. The mechanism works through the interaction of risk aversion and beliefs: we
show that an increase in an investor’s risk aversion leads to a desire to shift his portfolio
away from countries about which he is relatively optimistic, and towards those about
which he is relatively pessimistic.10
However, the effect of demand shifts on actual portfolio adjustments and asset prices
depend on the supply of assets faced by investors. We consider two polar cases. At one
extreme, we consider the case in which the supply of assets is completely inelastic. In this
case, the price of the assets adjusts so that, in equilibrium, total asset demand equals the
fixed asset supply. At the other extreme, we consider the case in which the supply of assets
is completely elastic. In this case, the quantities of assets adjust so that in equilibrium their
prices are constant.
Which assumption is more plausible empirically? The answer to that question is it
depends. In our empirical analysis, we use monthly data and focus on relatively high
frequency effects, in which case it may be more reasonable to assume that the supply
8 There exist several models of asset pricing in which investors agree to disagree, especially in the bubbles
literature. See for example Harrison and Kreps (1978), Scheinkman and Xiong (2003), and Hong and Stein
(2003). There may be other reasons why investors hold different portfolios. For example, countries may differ
in the volatility of dividends, or the correlation between a country’s dividends and investors’ marginal utility
may be different for different investors. We chose to assume differences in beliefs for simplicity.
10 We assume that there exists heterogeneity in investors’ beliefs and risk aversion, but investors are otherwise
similar. Other papers assume the existence of different classes of investors, but homogeneity within each class.
For example, Kodres and Pritsker (2002) assume the existence of informed investors, uninformed investors, and
noise traders, while Kyle and Xiong (2001) assume the existence of long-term value-based investors, convergence
traders, and noise traders.
9 To be able to solve the model analytically, we consider the case of CARA preferences and normally
distributed dividends, as in Calvo and Mendoza (2000), Kodres and Pritsker (2002), and Yuan (2004).
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230208
of assets is quite inelastic. However, the effective supply of assets faced by global
mutual funds may be increasing in the price they are willing to pay both because the
actual supply of assets may be somewhat elastic even in the short run, and also because
the demand by other investors on which we do not have data may be somewhat
elastic.11
1.1. Demand
There are two periods. There are two investors or fund managers, ia{�1,1}, whichpurchase assets in period 1 and consume in period 2. Investors can invest in three assets:
two countries, ca{�1,1}, and a safe asset. The countries pay stochastic dividends D�1and D1 in period 2 (and have zero residual value), and the safe asset has gross return 1.
Investor i’s utility is CARA with coefficient of absolute risk aversion ci.We assume that an investor values his own period 2 wealth WiVand also the difference
between his wealth and that of the other investor W�iV .
Ui ¼ � e�cid 1�að Þd WiVþad WiV�W�iVð Þð Þ for ia � 1; 1f g;
where a measures the degree to which investors care about relative returns as opposed toabsolute returns.12 In period 1, investors allocate their wealth Wi between each of the two
countries and the safe asset.
Investor i is relatively optimistic about country i and relatively pessimistic about
country � i. In particular, investor i believes Di ~N(DH,r2) and D� i ~N(DL,r2), whereDHNDL. The correlation between D�1 and D1 is 0.
13
We now calculate the demand for the two risky assets by each investor. Let Pc denote
the price of country c shares in period 1, and Xi,c the number of country c shares held by
investor i. Wealth levels in period 2 are thus given by
WiV ¼ Wi þXc
Xi; cd Dc � Pcð Þ for ia � 1; 1f g:
where Wi denotes initial wealth. Given the properties of CARA preferences and the fact
that returns are normally distributed, it is straightforward to obtain the four first order
conditions
BUi
BXi; c¼ 0 ¼ Ic¼id DH � Pc
� �þ Ic¼�id DL � Pc
� �� cid r2d Xi; c � ad X�i; c
� �; ð1Þ
for ia{�1,1} and ca{�1,1}. The first order conditions are easy to interpret. Otherthings equal, an investor prefers to invest in the country about which he is relatively
12 See Disyatat and Gelos (2001) for a discussion of this issue using a slightly different framework.13 Since we are only concerned with equilibrium in period 1, the actual probability distribution of the dividends
is irrelevant.
11 The papers by Calvo and Mendoza (2000) and Schinasi and Smith (2000) take returns as exogenous, which is
analogous to assuming a perfectly elastic supply of assets and exogenous prices. The papers by Kodres and
Pritsker (2002), Yuan (2004), and Kyle and Xiong (2001) assume a perfectly inelastic supply of assets, so that
returns and prices are endogenous but quantities are exogenous.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 209
optimistic. However, when an investor cares about relative returns (a N0), he has anincentive not to choose a portfolio very different from that of the other investor. This later
effect is relatively more important the more risk averse the investor is.
We now turn to the supply of assets. We consider the two polar cases of perfectly
inelastic supply (fixed quantities) and perfectly elastic supply (fixed prices).
1.2. Inelastic supply
Let the (fixed) number of country i shares be denoted by Ki. As a result, the market
clearing conditions for the two assets are
Kc ¼ X�1;c þ X1;c for ca � 1;1f g: ð2Þ
Eqs. (1) and (2) form a system of 6 linear equations and 6 unknowns: P�1, P1, X�1,�1,
X�1,1, X1,�1, X1,1. After some straightforward algebra, we get
Pc ¼DH þ DL
2� 1
c�1c þ c�1�c
� �d r2d 1� að Þd Kc
� cc � c�ccc þ c�c
� �d
DH � DL2
� �for ca � 1;1f g; and ð3Þ
Xi;c ¼Kc
2� ci � c�i
ci þ c�i
� �d
1� a1þ a
� �dKc
2
þ 1ci þ c�i
� �d
1
1þ a
� �d Ic¼id
DH � DLr2
� Ic¼�idDH � DL
r2
� �; ð4Þ
for ia{�1,1} and ca{�1,1}. Share prices in the two countries are equal to their averageexpected dividend, (DH+DL)/2, plus two additional terms. The first term is because, since
the assets are risky, they need to pay a premium for investors to hold them. This effect is
stronger the higher the variance of dividends r2, the higher the quantity of assets Kc, thehigher the levels of risk aversion ci, and the less investors care about relative returns (low a).
The second term is the most important result of the model. It shows that share prices
reflect the beliefs of the investor that is relatively less risk averse more than those of the
investor that is relatively more risk averse. In other words, if investor i is less risk averse
than investor � i the country about which investor i is relatively optimistic will tend tohave a higher price than the country about which investor i is relatively pessimistic. The
intuition is that very risk-averse investors tend not to act that much on their beliefs, so the
demand for the countries about which risk-averse investors are optimistic is relatively low.
Since the supply is inelastic, this lower demand is reflected in lower prices. The
transmission mechanism proposed in this paper hinges on this interaction between risk
aversion and beliefs.
With respect to asset allocations, each investor holds one half of each country’s shares,
Kc/2, plus two additional terms. The first term is because the less risk-averse investor will
tend to hold more of each of the two assets. The second term reflects the fact that each
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230210
investor will invest more in the country about which he is relatively optimistic and less in
the country about which he is relatively pessimistic.
Let bi,cuXi,c/(Xi,1+Xi,2) be investor i’s country c weight, defined as the share of totalinvestment in both countries that he invests in c. It is easy to show that, for all parameter
values, b1,1Nb�1,1 and b1,�1bb�1,�1. Namely, an investor tends to invest more than
other investors in the countries he is relatively optimistic about.
We can now describe how crises are transmitted across countries in this
environment. Assume that risk aversion depends on past performance, both absolute
and relative (positive a). Assume that there is a crisis in a third country in whichinvestor 1 is more heavily invested because he is relatively more optimistic about that
country. As a result of the crisis, investor 1 becomes more risk averse, both because he
suffered absolute losses and because his losses are higher than those of investor �1.Investor �1 may or may not become more risk averse but, even if he became morerisk averse, it would be to a lesser extent than investor 1. As a result, the crisis leads to
an increase in c1�c�1. From Eq. (3), we see that the price of country 1 shares would fallby a larger amount than those of country �1. The model suggests that the crisis should betransmitted to a larger extent to the country that shares optimistic investors with the crisis
country. Empirically, it is difficult to measure investor optimism. However, from Eq. (4)
we see that optimism is reflected in higher country exposures. As a result, the model
suggests that crises should affect to a greater extent countries that share overexposed
investors with the crisis country.
When the supply of assets is perfectly inelastic, portfolios do not adjust much as a result
of past performance. The reason is that while changes in risk aversion lead to changes in
asset demand, asset prices adjust so that investors end up holding the fixed quantity of
assets. To study the behavior of investors’ portfolios, we next study the case in which the
supply of assets is perfectly elastic.
1.3. Elastic supply
Let the (fixed) price of country i shares be denoted by P̄i. As a result, the market
clearing conditions (2) are replaced by
Pc ¼ P�c for ca � 1;1f g: ð5Þ
Replacing the prices in Eq. (1), we get a system of 4 linear equations and 4 unknowns:
X�1,�1, X�1,1, X1,�1, X1,1. After some straightforward algebra, we get
Xi;c¼1
r2d 1� a2ð Þd Ic¼idDH� P�c
ciþad D
L� P�cc�i
� �þIc¼�id
DL� P�cci
þad DH� P�cc�i
� �� �;
ð6Þ
for ia{�1,1} and ca{�1,1}. Two effects, reflected in the two terms in the secondfactor, drive each investor’s portfolio decisions. First, an investor wants to invest relatively
more in the country he is optimistic about. This effect is stronger the larger the difference
between the expected dividend (given his beliefs) and the country price and the lower his
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 211
level of risk aversion. Second, he wants to invest in the country where the other investor is
investing. This effect is stronger the higher the weight on relative performance a. Inaddition, the first factor shows that the lower the volatility of dividends and the more
investors care about relative performance, the more they invest in all countries.
We now turn to study the properties of country weights. In order for country weights to
be meaningful, we need to assume that the total investment in the two countries,
Xi ,�1+Xi ,1, is positive for both investors. It is easy to show that ((DH+DL)/
2� P̄1)+ ((DH+DL)/2� P̄2)N0 is a necessary and sufficient condition for this to be thecase. This condition is quite reasonable and it just states that the average country risk
premium is positive. As in the case of inelastic supply of assets, the fact that an investor
tends to invest more than other investors in the countries he is relatively optimistic about is
reflected in the fact that b1,1Nb�1,1 and b1,�1bb�1,�1 for all parameter values.
How do investors’ country weights respond to changes in risk aversion? In particular,
what is the effect of an increase in the risk aversion of investor i, ci, on the portfolio ofeach investor? It is easy to show that
dbi;i
dcib 0;
dbi;�idci
N 0;db�i;idci
b 0;db�i;�idci
N 0: ð7Þ
Namely, the investor whose risk aversion increases decreases his weight in the
country he is relatively optimistic about and increases it in the country he is relatively
pessimistic about. The other investor increases his weight in the country he is optimistic
about and decreases it in the country he is pessimistic about. The intuition behind these
results is straightforward. The increase in risk aversion makes investor i want to move
his portfolio closer to that of investor � i. Since bi,i Nb� i,i and bi,� i bb� i,� i, thisimplies a shift from country i to country � i. In turn, since investor � i’s countryweights also reflect an incentive not to have a portfolio very different from that of
investor i, he responds to the shift in investor i’s portfolio by shifting his own portfolio
in the same direction.
How do investors adjust their portfolios as a result of past performance? As in the
case of inelastic supply, we assume that risk aversion increases when investors’ past
performance is weak. As a result, weak past performance induces investors to move
towards the average investor’s portfolio by decreasing their exposure to countries in
which they were overexposed and increasing their exposure to countries in which they
were underexposed. In case of strong past performance, investors should adjust their
portfolios in the opposite direction. These effects are reinforced by a positive feedback
mechanism. If investor i suffers higher losses than investor � i, he should move towardsthe average portfolio. But investor � i should move away from the average portfolioboth because his relative performance was positive and also because the adjustment by
investor i shifts the average portfolio in the direction of investor � i’s portfolio. Thisadjustment by investor � i shifts the average portfolio away from investor i’s, whichgives investor i incentives to adjust his portfolio even further from his initial portfolio,
and so forth. As a result, we should expect relative performance to affect investors’
portfolios more than is suggested by the weight of relative performance in investors’
utilities a.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230212
1.4. Some comments on the model
The transmission mechanism we propose in this paper consists of three steps: (i)
weak relative performance due to an overexposure to the crisis country increases
investors’ risk aversion; (ii) the increase in risk aversion, in turn, produces a
retrenchment towards the average portfolio; and (iii) the retrenchment of overexposed
investors, in turn, leading to a drop in stock prices in countries that share overexposed
investors with the crisis country. Indeed, this explanation de-emphasizes the role of
macroeconomic fundamentals in explaining which countries are impacted by the crisis
or the so-called bwake-up hypothesisQ in which investors bwake-upQ and begin to takenote of other countries’ similarities with the original crisis country.14 The model
illustrates steps (ii) and (iii). However, the transmission mechanism depends on the
assumption that risk aversion increases with weak relative performance, needed for step
(i). Although we cannot directly test this assumption, it is consistent with the behavior of
investors’ portfolios, as shown below. In addition, this assumption is consistent with some
findings in the finance literature that analyze the behavior of mutual fund managers:
inflows and outflows out of mutual funds depend to a large extent on the relative
performance of funds, and fund managers’ income increases with assets under
management.15 In our model, investors’ portfolios do not display wealth effects due to
our use of CARA preferences. Under more realistic preferences that display diminishing
absolute risk aversion (such as CRRA), a decrease in wealth due to past losses and a
withdrawal of funds would tend to decrease risk taking by investors without needing to
assume changes in their utility function. We did not study such preferences because CARA
preferences are necessary to obtain closed-form solutions. However, the results from a
model with diminishing absolute risk aversion and wealth effects would be qualitatively
similar to our results.16
In our analysis, we study separately the cases of perfectly inelastic and perfectly elastic
asset supply. Under perfectly inelastic supply we examine the transmission of crises, while
under perfectly elastic supply we analyze the behavior of investors’ portfolios. In reality,
the supply of assets faced by global mutual funds is likely neither perfectly elastic nor
perfectly inelastic. In such a case, both sets of results would hold, although the effects
would be quantitatively smaller.
In the following sections we examine both the behavior of investors’ portfolios and the
transmission of shocks. We study whether poor past performance leads investors to
14 The term bwake-up hypothesisQ was coined by Goldstein (1998).15 See Appendix A for a more complete discussion of these findings and for some complementary evidence
based on our data. Related assumptions are also often used in the literature on herding; see, for example,
Scharfstein and Stein (1990).16 Our mechanism is related to the one in Shleifer and Vishny (1997). They also assume that past relative
performance affects funds under management by investors (in their case, by arbitrageurs). As a result, in their
model weak relative performance leads to a decrease in the amount of funds investors can invest in assets they are
relatively more optimistic about, due to financial constraints. In our model, investors decrease their exposure to
countries about which they are relative more optimistic because we further assume that the decrease in funds
under management increases risk aversion. As mentioned above, this last assumption would not be necessary
under diminishing absolute risk aversion.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 213
bretrenchQ towards the average portfolio. (We do so by regressing changes in countryweights on the interaction of past performance and country overexposure.) We then
examine whether a crisis in one country is transmitted to a greater extent to countries that
share overexposed investors with the crisis country — this is done by constructing an
index of financial exposure that reflects such common-investor links, and testing the
power of this financial exposure index in explaining stock market comovements and in
predicting three recent crises episodes.
2. Data
The mutual fund data used in this paper are from a comprehensive database
purchased from eMergingPortfolio.com. The database covers (on a monthly basis),
the geographic asset allocation of hundreds of equity funds with a focus on
emerging markets for the period 1996:1–2000:12. The funds are domiciled in
different countries around the world. At the beginning of the sample, the database
contains 382 funds with assets totaling US$117 billion. At the end of the sample,
the number of funds increased to 639, with US$120 billion in assets. While the
total number of funds rose over the period, some funds were dropped from the
database if they discontinued providing information on their holdings. We focus on
global dedicated emerging funds, i.e. funds that invest in emerging markets
worldwide.17 For stock market returns, we used monthly IFC US$ total returns for the
period 1990–2000, complementing them whenever needed with data from MSCI or
national sources.
In December 2000, the subsample consisted of 117 global emerging market funds.
Approximately one quarter of the funds are closed-end funds. The assets of these funds
represent a modest, but not negligible fraction of the total market capitalization in the
countries they invest. For example, in the case of Argentina, funds held approximately
2.7% of the total stock market capitalization in August of 1998, while the share was
around 1.3% for Korea.
While precise numbers on total equity flows are hard to obtain, a substantial fraction of
all equity flows to emerging markets seems to occur through the funds in our database.
According to the World Bank (2003), in 1998, total portfolio equity flows to developing
countries amounted to US$7.4 billion, compared to US$ 0.8 billion flows recorded in our
sample.
The company providing this data aims for the widest coverage possible of
emerging market funds without applying any selection criteria. According to the
provider, the complete database covers roughly 80% of all dedicated emerging market
funds, with coverage of about 90% of total emerging market fund assets. We do not
have data on holdings of individual stocks or on the timing of funds’ purchases and
sales over the month. We calculate the implied flows from the asset position data,
17 For more details on the data, see Borensztein and Gelos (2003). Kaminsky, Lyons and Schmukler (2004) also
examine mutual fund behavior in emerging markets worldwide but use data at a more aggregate level.
http:eMergingPortfolio.com
-
Distance from Average Portfolio and CumulatedAverage Fund Returns
0
20
40
60
80
100
120
9601
9606
9611
9704
9709
9802
9807
9812
9905
9910
2000
3
2000
8
Cum
ulat
ed m
ean
retu
rns
0
5
10
15
20
25
30
35
Med
ian
dist
ance
fro
m m
ean
port
folio
Cum. Ret.
Median Dist.
Fig. 1. Portfolio dispersion over time. Distance from average portfolio is the median portfolio distance from the
mean portfolio. The distance is measured as the root mean squared difference over country weights. These
averages are based on global funds only.
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230214
assuming that within countries, funds hold a portfolio that is well proxied by the IFC
US$ total return investable index.18 We also assume that flows occur halfway through the
month.
3. Portfolio dispersion over time
To obtain a first impression of the data, in this section we compute the dispersion of
fund portfolios over time. We measure dispersion as the root mean squared distance over
country weights between each fund and the average portfolio, where the average
portfolio is weighted by fund size. Fig. 1 shows the median of this dispersion for the
group of global funds, together with the cumulated mean fund returns (set equal to 100
at the beginning of the sample). The picture shows that fund portfolios started
converging during the Asian crises – possibly back to more normal levels after the run
up – at the same time that funds started facing large portfolio losses. This suggests that
during turbulent times, funds tend to retrench towards the average. However,
improvements in performance after the Russian crisis were not accompanied by an
increase in fund dispersion. In the next section, we examine in detail how fund portfolio
choices depend on their performance. We show that funds do retrench towards the mean
during periods of low returns, but they react to returns relative to those of other funds as
opposed to absolute returns. This distinction has important implications for the
transmission of shocks during crises, since relative returns are very sensitive to whether
funds are overexposed to crisis centers.
18 This turns out to be a good approximation in emerging markets. See Borensztein and Gelos (2003).
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 215
4. Fund performance and portfolio choice
This section analyzes the trading behavior of emerging market mutual funds. We
concentrate on the effect of portfolio returns – both absolute and relative to the average
portfolio – on funds’ portfolio decisions. For this purpose, we regress changes in portfolio
weights (one observation per fund–country–date) on overexposure, excess gains (or
losses), gains, and the interactions of excess gains and gains with overexposure. We find
that, as predicted by the model, when fund returns are lower than that of the average
portfolio, funds reduce their exposure to countries in which they were boverweightQ andincrease their exposure to countries in which they were bunderweight.Q
Let sub-indices i denote fund, c country, and t time. Let ai,c,t denote assets and rc,t the
stock index return. Let si,t =P
cai,c,t denote the size of a mutual fund, bi,c,t =ai,c,t/si,t its
country weight, and b̄c,t the average (weighted by fund size) country weights across funds.
Let overexposure oei,c,t, fund gains gi,t, and fund excess gains exgi,t be defined as
oei;c;t ¼ bi;c;t � b�c;t;
gi;t ¼Xc
bi;c;t�1rc;t;
exgi;t ¼ gi;t �Xc
b�c;t�1rc;t:
The change in country weight, dbi,c,t, is given by
dbi;c;t ¼ bi;c;t � bi;c;t�1:
It is not clear that we should focus on dbi,c,t as a measure of portfolio adjustment by
funds. For example, if the market capitalization of a country as a fraction of total world
market capitalization changed, one would expect that, on average, mutual funds’ country
weights would adjust as well. In particular, it is obvious that it would not be possible for all
investors (mutual funds and others) to keep a constant country weight.
At one extreme, if the supply of assets were totally inelastic market capitalization would
change proportionately to country returns rc,t.19 As a result, even if funds acted passively
without buying or selling shares, the country weight would change by an amount
adji;c;t ¼bi;c;t�11þ gi;t
� �rc;t � gi;t� �
:
In this case, one would want to use an badjustedQ change in weights, dbi,c,tV , that solelycaptured the change in weights that arose from funds actively buying and selling assets,
dbi;c;tV ¼ dbi;c;t � adji;c;t: ð8Þ
From the discussion in Section 1.2, we see that share prices and expected returns would
adjust in order to keep investors content holding the resulting portfolio. For example, if in
one country returns are lower than average, we would expect share prices not to fall
19 This would not be exactly true if firms paid dividends. However, at monthly frequencies dividends are not an
important fraction of returns, especially for emerging markets.
-
Table 1a
Portfolio adjustment: 1996:1–2000:12
Assuming inelastic
supply (bu1)Assuming perfectly
elastic supply (bu0)No assumption on
supply elasticity
(b unconstrained)
Adjustment term 1 0 0.436***
(0.007)
Overexposure (t�1) �0.044***(0.002)
�0.069***(0.002)
�0.061***(0.002)
Excess gains 3.360***
(0.23)
�0.831***(0.244)
1.045***
(0.233)
Overexposure (t�1)�excess gains 0.647***(0.092)
1.035***
(0.094)
0.870***
(0.090)
R2 0.02 0.02 0.12
Observations 40,946 38,353 38,353
Dependent variable: change in country weight, as defined in Eq. (8). This analysis is based on one observation per
fund–time–country. All variables normalized by beginning of period fund size. ***, **, and * represents
statistical significance at the 1%, 5%, and 10% level, respectively. Numbers in parenthesis are standard errors.
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230216
proportionately as much as expected dividends, since the expected returns need to fall to
keep investors from wanting to reestablish their prior country weights.
At the other extreme, if the supply of assets were totally elastic expected returns would
remain constant and, thus, we would expect funds to keep constant country weights. In this
case, one would want to use the unadjusted change in weights, dbi,c,t, in the regressions.
Finally, for intermediate cases one would want to adjust dbi,c,t, but by less than in Eq. (8).
We run the following regression
dbi;c;t ¼ ad oei;c;t�1 þ bd adji;c;t þ cd exgi;t þ dd oei;c;t�1d exgi;t þ ei;c;t:
The first term captures possible mean reversion in portfolios. The role of the second term
should be clear from the discussion above. We run three types of regressions: one cons-
training b to be 1 which corresponds to the case of perfectly inelastic supply, one cons-training b to be 0 which corresponds to the case of perfectly elastic supply, and one in whichb is unconstrained, letting the regression tell us what the appropriate adjustment term is.
If our hypothesis were true, fund i should increase its weight on country c (dbi,c,tpositive) if the fund was overexposed to country c (oei,c,�1 positive) when the fund is
doing relatively well (exgi,t positive). Likewise, the fund should increase its weight on
country c (dbi,c,t positive) if the fund was underexposed to country c (oei,c,t�1 negative)
when the fund is doing relatively badly (exgi,t negative). As a result, we focus on the
coefficient d, which should be positive according to our hypothesis.Funds indeed tend to buy into countries in which they are overexposed (underexposed)
when their gains are higher (lower) than that of other funds. Tables 1a and 1b summarize our
results for the three cases in which bu1, bu0, and b is unconstrained.20 We report results
20 We restricted the sample to countries that represent at least 1% of average fund portfolio. We observed that we
could explain portfolio adjustments for large countries better than for small countries. One possible explanation is
that the index is mismeasured for small countries due to rounding off in portfolio reporting by funds. The raw data
indeed seems to be rounded.
-
Table 1b
Portfolio adjustment: 1996:1–2000:12
Assuming
inelastic
supply
Assuming
inelastic
supply
Assuming
perfectly
elastic supply
Assuming
perfectly
elastic supply
No restriction
on supply
elasticity
No restriction
on supply
elasticity
Adjustment term 1 1 0 0 0.435***
(0.007)
0.433***
(0.007)
Overexposure (t�1) �0.044***(0.002)
�0.045***(0.002)
�0.069***(0.002)
�0.070***(0.002)
�0.061***(0.002)
�0.061***(0.002)
Excess gains 3.289***
(0.253)
– �1.568***(0.262)
– 0.588**
(0.251)
–
Overexposure (t�1)�excess gains 0.843***(0.100)
– 1.211***
(0.102)
– 1.062***
(0.097)
–
Gains 0.106*
(0.078)
0.487***
(0.072)
0.651***
(0.080)
0.488***
(0.074)
0.420***
(0.076)
0.500***
(0.071)
Overexposure (t�1)�gains �0.173***(0.034)
�0.053***(0.031)
�0.179***(0.034)
�0.028(0.032)
�0.182***(0.032)
�0.043(0.030)
R2 0.02 0.01 0.03 0.02 0.12 0.12
Observations 40,946 40,946 38,353 38,353 38,353 38,353
Dependent variable: change in country weight, as defined in Eq. (8). This analysis is based on one observation per fund–time–country. All variables normalized by
beginning of period fund size. ***, **, and * represents statistical significance at the 1%, 5%, and 10% level, respectively. Numbers in parenthesis are standard errors.
F.A.Broner
etal./JournalofIntern
atio
nalEconomics
69(2006)203–230
217
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230218
including excess gains as well as gains to determine whether funds care more about
relative or absolute performance. In all cases, the coefficient d is positive and statisticallysignificant at the 1% level. There is also a significant reversion to the mean in the sense
that on average funds buy into countries were they are underexposed. It is interesting to
note that excess gains seem to be more important than absolute gains, both in levels and
when interacted with overexposure. When including absolute gains, the interaction term of
lagged overexposure and absolute gains is small and not always significant. Finally, when
unconstrained, the coefficient on the adjustment term is always significantly greater than 0
and significantly lower than 1, suggesting that indeed mutual funds face neither a perfectly
elastic nor a perfectly inelastic supply of assets.
The economic significance of the effect of funds’ relative performance on whether or
not they retrench to the benchmark is moderate, but by no means negligible. For
example, consider a country in which half the funds (weighted by fund size) invest 15%
of their assets and half the funds invest 5% of their assets, so that the former have
overexposure of +5% and the latter of �5%. Assume that the first group of funds has
Table 1c
Portfolio adjustment with control variables: 1996:1–2000:12
Assuming inelastic
supply
Assuming perfectly
elastic supply
No restriction on
supply elasticity
Adjustment term – – 0.427**
(0.007)
Overexposure (t�1) �0.044***(0.002)
�0.070***(0.002)
�0.061***(0.002)
Excess gains 3.415***
(0.237)
�0.814***(0.245)
1.045***
(0.235)
Overexposure (t�1)�excess gains 0.665***(0.094)
1.023***
(0.095)
0.872***
(0.091)
DEconomic risk (t�1) �0.005(0.004)
0.014***
(0.004)
0.006***
(0.004)
DFinancial risk (t�1) �0.005(0.004)
�0.009***(0.004)
�0.007***(0.004)
DPolitical risk (t�1) 0.006**(0.003)
0.005
(0.003)
0.005**
(0.003)
Overexposure (t�1)�Deconomic risk (t�1) �0.002(0.002)
0.001
(0.002)
�0.001(0.002)
Overexposure (t�1)�Dfinancial risk (t�1) 0.003*(0.002)
0.007***
(0.002)
0.006***
(0.002)
Overexposure (t�1)�Dpolitical risk (t�1) 0.000(0.001)
�0.002(0.001)
�0.001(0.001)
R2 0.02 0.03 0.11
Observations 39,691 37,174 37,174
Dependent variable: change in country weight. This analysis is based on one observation per fund–time–country.
All variables normalized by beginning of period fund size. ***, **, and * represents statistical significance at the
1%, 5%, and 10% level, respectively. The Ds denote first differences. Economic, financial, and political risk referto the International Country Risk Guide’s (ICRG) monthly economic, financial, and political risk indices.
Numbers in parenthesis are standard errors.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 219
losses of 10% while the second group has gains of 10. According to the results in Table
1a (unconstrained b), both groups of funds would reduce their weight in the country by0.44%. In addition, the first group of funds will now manage 0.5 d 90% of total fund
assets while the second group of funds will correspond to 0.5 d 110% of total fund
assets. As a result, the average weight of the country in total fund assets would drop
from 10% to 9.07%, which implies that total funds’ investment in the country would
drop by almost 10%. In addition, the 10% drop in funds’ investment in the country
would take place despite the fact that the expected returns in the country would have
increased, since the supply of assets is not perfectly elastic.
We also ran regressions including control variables. There, we added variables such as
changes in risk as reported by the International Country Risk Guide (ICRG); we included
such control variables independently and as interactions with lagged excess gains.21 Table
1c shows the results of adding to our base regression changes in economic risk, financial
risk, and political risk (all from ICRG), both in levels and when interacted with
overexposure. The table shows that these variables do not have any effect on the estimates
of the coefficients of interest shown in Tables 1a and 1b. Furthermore, these risk measures
are not statistically significant.
5. An index of financial interdependence
The results in the previous section suggest that the effect of crises on fund flows
depends systematically on funds’ degree of overexposure to the crisis country. In
particular, since the funds that were overexposed to the crisis country are likely to have
larger losses than those that were underexposed, we should expect those funds to take
capital out of the countries in which they were overexposed and into the countries in which
they were underexposed.
In this section we construct a matrix or index of bfinancial interdependenceQ betweencountries based on whether countries share overexposed fund investors. We define country
c1’s reliance on fund i, rec1i,t, as the contribution of fund i to total investment in the
country by all funds,
rec1;i;t ¼ai;c1;tX
i V
ai V;c1;t:
We define country c1’s reliance on investors overexposed to country c2, dc1,c2,t, as
dc1;c2;t ¼Xi
rec1;i;t � oei;c2;t;
namely, the sum of every fund’s overexposure to country c2, weighted by c1’s reliance on
each fund. For short, we also refer to dc1,c2,t as country c1’s exposure to country c2. The
21 In an earlier version, we also looked at the differences between open-end and closed-end funds. We found that
the two types of funds behave similarly. The coefficient d was always positive and significant at the 1% level, andits magnitude was slightly higher for closed-end funds.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230220
relationship between this definition of exposure and the results in Section 4 can be
illustrated by noting that dc1,c2,t can be rewritten as
dc1;c2;t ¼Xi
si
S
oei;c1;t
b�c1;t
� oei;c2;t;
where S=P
isi, the sum of the assets of all funds. (See Appendix B for details.) As shown
in Section 4, a fund should reduce its investments in country c1 in response to low excess
gains if that fund is overexposed to country c1. This explains why the exposure measure is
related to the correlation between funds overexposure to the crisis country and to c1. The
reason why oei,c1,t is divided by b̄c1,t is that the effect of a given reduction in funds
investments in country c1 will depend on the size of that reduction relative to total
investments in the country. That is why the exposure measure is not symmetric,
dc1,c2,t p dc2,c1,t. Note that this does not mean that small countries are, in general, moreexposed to crises, since funds overexposure to small countries tends to be small. On the
other hand, it is true that countries, in general, have low exposures to small countries.
6. Financial interdependence and contagion
If the financial exposure index developed here contains useful information about
financial linkages across countries, one would expect it to help explain comovement
patterns across stock markets. In particular, while a country’s stock market return is likely
to be influenced by those of other countries through a variety of channels, our results so far
suggest that sharing common overexposed investors with another market will increase the
transmission of shocks from that market (Table 2). To address this question, we run the
following regression over the whole sample period,
rc;t ¼ aþ bdXcVp c
wcVd rcV;t þ cdXcVp c
tc;cVd rcV;t þ ddXcVp c
dc;cV;t�1d rcV;t þ ec;t;
where rc,t is country c’s stock return at time t, wc is country c’s stock market size (as
measured by the IFC investable index), tc,cVis the trade share of country cVin country c’s
Table 2
Transmission channels: 1996:2–2000:12
(1) (2) (3) (4) (5) (6)
Stock market size 0.0087***
(0.0005)
0.0087***
(0.0006)
0.0085***
(0.0006)
Trade linkages 0.8522***
(0.2561)
0.8932***
(0.2631)
0.1325
(0.1772)
Financial exposure 0.00077***
(0.00017)
0.00064***
(0.000123)
0.00084***
0.00015)
0.00064***
(0.00012)
R2 0.24 0.03 0.03 0.26 0.06 0.26
Observations 990 1079 1108 929 1049 880
Dependent variable: monthly country stock market returns. Robust standard errors are in parentheses. ***, **,
and * represents statistical significance at the 1%, 5%, and 10% level, respectively.
-
Table 3a
Stock market returns during three crises: Thailand 1997, Russia 1998, and Brazil 1999
Thailand Russia Brazil
Weight N1 Weight N2 Weight N1 Weight N2 Weight N1 Weight N2
Financial exposure �0.368***(0.124)
�0.504***(0.093)
�0.081**(0.033)
�0.057*(0.031)
�0.021*(0.012)
�0.039*(0.020)
R2 0.28 0.52 0.15 0.09 0.08 0.20
Observations 19 14 19 15 21 14
Stock market returns as a function of a country’s exposure to crisis countries. The Thai crisis regression
corresponds to cumulative returns during April 1997–August 1997, the Russian crisis regression to July 1998–
September 1998, and the Brazilian crisis regression to January 1999. Weight refers to the minimum weight of a
country in the average portfolio to be included in regressions. Exposure variable lagged one from beginning of
crisis. Crisis countries excluded from regressions. ***, **, and * means statistical significance at the 1%, 5%, and
10% level, respectively. Robust standard errors are shown in parentheses.
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 221
total trade, and dc,cV,t is the financial exposure of country c to country cV, as defined earlier.In other words, we assess whether the extent to which country c’s stock returns are
influenced by those in country cVdepends on the size of the stock market of country cV, thetrade linkages between countries c and cV, and country c’s financial exposure to country cVthrough common investors.
Our index of financial exposure does help explain country’s stock returns. In both
simple OLS and country-fixed effects regressions, d enters significantly with a positivesign. This result is robust to the inclusion of lagged dependent and independent
variables. When all three – the size-weighted, trade-weighted, and financial-links-
weighted returns of the rest of the emerging market world are included – the trade-
weighted returns become insignificant, suggesting that financial linkages as identified
here may be more important in the transmission of shocks across stock markets than
trade connections. Given that we are examining only a subset of international investors,
such a positive finding could be interpreted as an indication that mutual funds are
representative of international investors in general. However, a causal interpretation
remains difficult. To address this aspect more directly, we now examine the impact of
crises.
Does our index of financial exposure help predict which countries are likely to be
affected by contagion? We now take a closer look at the pattern of cross-country stock
market movements during the Thai, Russian and Brazilian crises (Table 3a).22 For the
22 The crisis dates were chosen as follows: In Thailand, difficulties were apparent since the beginning of 1997,
the currency was devalued in June, and the biggest drop in the stock market took place in August. As a result, for
the Thai crisis we study accumulated stock market returns during the period April 1997–August 1997. In Russia,
interest rates on T-bills increased substantially in July 1998, the default took place in August, and the large drops
in the stock market took place in August and September. As a result, for the Russian crisis we study accumulated
stock market returns during the period July 1998–September 1998. In Brazil, it is difficult to pinpoint to a start of
the crisis, as pressure started mounting beginning with the Russian default. As a result, for the Brazilian crisis we
study the returns during January 1999, the month when both the devaluation and the largest stock market drop
took place.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230222
three crises, we run separate regressions of stock market returns on exposure restricting the
sample to countries that represent at least 1% and 2% of the average fund portfolio,
respectively.23
For all crises, the coefficient on the financial exposure variable is negative and
statistically significant, albeit not always at high confidence levels. For the Thai crisis, the
financial exposure variable is significant at the 1% level. Furthermore, the exposure
variable explains between 28% and 52% of the cross-sectional variation in country returns.
For the Russian crisis, the financial exposure variable is significant at the 5% level and
explains 15% of the cross-sectional variation in country returns. For the Brazilian crisis,
the financial exposure variable is significant at the 10% level for countries with weights
greater than 1%. In addition, both significance and explanatory power increase, as the
sample is restricted to larger countries. For countries with weights greater than 3% (not
shown), the exposure variable explains 45% of the cross-sectional variation in stock
returns.
Fig. 2 illustrates the effect of exposure on returns, restricting the sample to countries
with weights greater than 1%. First, it seems clear that the results are not due to outliers.24
Second, it shows that focusing on financial exposure, we can explain why some countries
with no other obvious links to the crisis country suffered contagion, while others that ex
ante might have seemed connected did not. During the Thai crisis, among the Asian
countries Taiwan was relatively unaffected, perhaps because it did not share overexposed
investors with Thailand. Malaysia, on the other hand, was the country most affected and
also the country most exposed. During the Brazilian crisis, Argentina was the country most
exposed and also one of the 3 with lowest returns and the lowest among Latin-American
countries. In addition, both among European countries and among Asian countries, those
with high exposure had lower returns than those with low exposure (China being the
exception).
Next, we examine the importance of three important control variables, adding them
one at a time to each regression (Table 3b). First, the presence of trade linkages is an
important potential candidate for explaining the pattern of financial shock comovements
across countries. Therefore, we include an index of the degree of direct trade
competition as used in Van Rijckeghem and Weder (2001). Second, we use two
variables measuring the degree to which country i competes for funding from the same
bank lenders as the crisis country, as proposed by Kaminsky and Reinhart (2000) and
Van Rijckeghem and Weder (2001).25 The first of these indices is based on the absolute
value of credits obtained from the common lender, and the second is based on the share of
23 We observed that the index of financial interdependence explains returns in large countries better than in small
countries. This parallels our finding that portfolio adjustments could also be explained for large countries better
than for small countries.24 One might suspect that Malaysia could be such an outlier in Fig. 2. When removing Malaysia from the
sample, the standard errors in the first column of Table 3a do become substantially larger. However, the
coefficients remain highly significant in the second column.25 See bfunds competitionQ in Table 1, p. 300 in Van Rijckeghem and Weder (2001). We are grateful to the
authors for sharing their data with us. (The Brazil crisis was not covered in their study and we constructed the data
for this case.)
-
Thai crisis
retu
rns
exposure to crisis country-1 -.5 0 .5 1
-.5
0
.5
argent
brazilchile
china
greece
hongkoindia
indone
koreas
malays
mexico
peru
philip
poland
russia
safric
singap
taiwan
turkey
Russian crisis
retu
rns
exposure to crisis country-2 -1 0 1
-.6
-.4
-.2
0
.2
argent
brazil
chile
greece
hongko
hungar
india
indone
koreas
malays
mexico
peruphilip poland
safric
singaptaiwan
thaila
turkey
Brazilian crisis
retu
rns
exposure to crisis country-3 -2 -1 0 1
-.2
0
.2
argent
chile
china
greece
hongko
hungar
india
indone
israelkoreas
malays
mexicoperu
philip
poland
russia
safric
singap
taiwan
thaila
turkey
Fig. 2. Exposure to crisis country and stock market returns.
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 223
-
Table 3b
Stock market returns during crises, including control variables
Thailand
Financial exposure
(lagged)
�0.368***(0.124)
– – – – �0.324***(0.106)
�0.378***(0.088)
�0.406***(0.115)
�0.341***(0.123)
Trade competition – �0.551(0.387)
– – – �0.366(0.275)
– – –
Competition for bank
funds (share)
– – �0.608(0.503)
– – – 0.039
(0.458)
– –
Competition for bank
funds (absolute)
– – – 0.319
(0.334)
– – – 0.472
(0.328)
–
ICRG financial risk
index (lagged)
– – – – �0.015(0.011)
– – – �0.010(0.007)
R2 0.28 0.13 0.11 0.04 0.08 0.33 0.28 0.37 0.31
Observations 19 19 19 19 19 19 19 19 19
Russia
Financial exposure
(lagged)
�0.081**(0.0334)
– – – – �0.039(0.043)
�0.023(0.039)
�0.084**(0.033)
�0.074*(0.040)
Trade competition – �3.996**(1.682)
– – – �3.537*(1.923)
– – –
Competition for bank
funds (share)
– – �0.827***(0.255)
– – – �0.732*(0.362)
– –
Competition for bank
funds (absolute)
– – – �0.096(0.254)
– – – �0.150(0.239)
–
ICRG financial risk
index (lagged)
– – – – 0.005
(0.006)
– – – 0.004
(0.006)
R2 0.15 0.18 0.30 0.01 0.05 0.20 0.32 0.17 0.21
Observations 19 18 19 19 18 18 19 19 19
F.A.Broner
etal./JournalofIntern
atio
nalEconomics
69(2006)203–230
224
-
Brazil
Financial exposure
(lagged)
�0.021*(0.021)
– – – – �0.016(0.012)
�0.028*(0.015)
�0.025(0.019)
�0.021*(0.011)
Trade competition – �0.713***(0.200)
– – – �0.581**(0.250)
– – –
Competition for bank
funds (share)
– – 0.001
(0.137)
– – – 0.013
(0.138)
– –
Competition for
bank funds (absolute)
– – – �0.095(0.209)
– – – 0.078
(0.284)
–
ICRG financial risk
index (lagged)
– – – – 0.001
(0.003)
– – – 0.000
(0.003)
R2 0.08 0.10 0.00 0.01 0.08 0.14 0.10 0.08 0.08
Observations 21 21 21 21 21 21 21 21 21
Stock market returns as a function of a country’s exposure to crisis countries. The regressions correspond to cumulative returns during April 1997–August 1997 for the
Thai crisis, July 1998–September 1998 for the Russian crisis, and January 1999 for the Brazilian crisis. We only included countries with an average weight in fund
portfolios of at least 1%. Exposure variable lagged one from beginning of crisis. Crisis countries excluded from regressions. ***, **, and * means statistical significance at
the 1%, 5%, and 10% level, respectively. Robust standard errors are shown in parentheses. For the variables btrade competitionQ and bcompetition for bank fundsQ see VanRijckeghem and Weder (2001).
F.A.Broner
etal./JournalofIntern
atio
nalEconomics
69(2006)203–230
225
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230226
borrowing from the common lender. Third, we use the financial risk score from ICRG.
Due to the limited number of observations, we cannot include lists of potentially relevant
macroeconomic fundamentals.26
The small number of observations limits inference but some patterns are observable.
For the Thai crisis, none of the control variables are significant in explaining the
pattern of the stock market reaction across countries, and the coefficient on our
financial exposure variable remains broadly unchanged and statistically significant
when including the control variables. For the Russian crisis, the trade variable is
significant and alone explains a similar share of the total variance in stock returns. The
babsoluteQ bank competition variable used by Van Rijckeghem and Weder (2001) forthe Russian crisis does not enter significantly. However, the bshareQ-based bankcompetition variable is significant, and explains 30% of stock returns variation. When
including both the financial exposure index and one of the three control variables at a
time, the results are mixed. The financial exposure index becomes insignificant when
including the trade competition variable or the bshareQ-based bank competition index.The financial exposure variable, does however, survive the inclusion of the babsoluteQbank competition index and the inclusion of the ICRG financial risk index. For the
Brazil crisis, the pattern is similar: trade linkages matter, and the financial
interdependence variable remains statistically significant when controlling for bank
linkages (which do not seem to matter) and financial risk (insignificant) but becomes
insignificant when adding trade competition.27 However, all these results must be
interpreted with care, as restricting our sample to the crises episode severely reduces the
number of observations.
7. Conclusions
We have shown that the tendency of mutual funds to respond to relative losses
(gains) by moving closer (further away) to (from) the average portfolio helps in
explaining the transmission of shocks across countries. This behavior may exacerbate
the effect of crises, by creating both contagion between countries and momentum
trading at the country level. This, in turn, prompts the question of whether countries
should limit participation of international funds in their stock markets to index funds
(i.e., funds that passively follow the index). However, there is reason to believe that
such a measure would likely be counterproductive. Information gathering by investors
such as emerging market funds plays a useful role, and if all investors blindly followed
27 Johnson et al. (2000) have argued that corporate governance indices can help explain the pattern of stock
market declines during the Asian crisis. In a related vein, Gelos and Wei (in press) show that funds tend to avoid
intransparent countries during crises. We did not investigate this issue here.
26 We experimented with probabilities of currency crises as predicted by the early warning system used at the
IMF and described in Berg and Pattillo (1999). This variable summarizes the information contained in a
variety of macroeconomic variables. However, it is only available for a subset of countries in our sample,
reducing our sample size further. When included, the variable was never significant at the 5% confidence
level.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 227
indices, the indices themselves might become arbitrary, yielding herding in an extreme
form.28
The predictive power of our index of financial exposure based on international mutual
funds we conjecture may reflect the fact that these funds are representative of other kinds
of investors, such as commercial banks and investment houses. In order to gain a more
complete picture of the functioning of international capital markets, however, this strand of
research can be complemented by a similar examination of other market players’ behavior.
In particular, it would be useful to gain insight into the portfolio decisions of international
institutional investors investing in fixed-income securities, both in local as well as in
international markets. While data remain scarce in this area, some data sources are
becoming available, offering opportunities for deepening our understanding of the
vagaries of international financial flows — particularly in times of market stress.
Appendix A. Fund performance and redemptions
The relationship between relative performance, inflows and outflows of assets, and risk
taking by mutual funds has been studied previously in the finance literature, especially in the
context of US domestic funds. There is clear evidence that mutual funds with weak relative
performance suffer withdrawals of assets relative to better performing funds (see Ippolito,
1992). It is also clear that mutual funds earnings are increasing in, and approximately
proportional to, the amount of assets under management, with fees around 0.5% of assets.
With respect to the relationship between performance and risk taking, the evidence is less
clear. Initial studies pointed to the presence of bgambling behaviorQ by fund managers whofall behind in their performance (see Brown et al., 1996; Chevalier and Ellison, 1997; Sirri
and Tufano, 1998). The reason is that mutual funds with the best performance capture the
lion’s share of new inflows while funds that perform poorly are not penalized equally, so
managers seem to have incentives to choose more risky portfolios if they are falling behind.
More recent studies, however, have questioned this result. Busse (2001) finds that mid-year
losers decrease their risk during the second half of the year. Koski and Pontiff (1999) report a
positive correlation between current risk taking and past-year performance. Daniel and
Wermers (2000) find that prior risk-taking behavior is a much better predictor than prior
performance in explaining future risk-taking behavior by fund managers. Chen and
Pennachi (2002) argue that while fund managers do increase a fund’s tracking error as its
relative performance declines, this does not result in a higher variance of the fund’s returns.
To our knowledge, ours is the first paper that studies the effect of performance on risk
taking by dedicated emerging market funds. Given our short sample and the fact that we
do not have precise data about inflows of assets, we cannot conduct an analysis
comparable to those above. Instead, we carried out the following crude exercise. First, we
constructed data on inflows and outflows of assets for global funds indirectly, subtracting
imputed fund gains from increases in reported size, and dividing by fund size. Then, we
28 This point has been made by Calvo and Mendoza (2000). More generally, this question touches on one of the
paradoxes of the efficient market hypothesis: if markets are efficient, it does not pay to gather information, but
markets cannot be efficient if nobody bothers to gather information. See Grossman and Stiglitz (1980).
-
.04
.042
.044
.046
Excess inflows In first quarter
-.3 -.2 -.1 0 .1 .2
Past year’s excess return
Fig. A1. Fund performance and inflows. Local polynomial regression of excess inflows (in excess of average
inflows across funds) in the first quarter of a year on past year’s excess return (in excess of average fund returns).
The estimation uses an Epanechnikov kernel with a width of 0.3.
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230228
estimated nonparametrically the relationship between excess inflows in a given quarter and
past year’s excess returns. The relationship, plotted in Fig. A1, seems to be positive, but
there is no evidence that funds gain more by being top performers than they lose by being
worst performers.
Appendix B. Equivalence of indices of interdependence
We start from the first index of financial interdependence
dc1; c2 ¼Xi
rec1; i � oei; c2 ;
where we have removed time sub-indices for simplicity. This expression can be rewritten as
dc1;c2 ¼Xi
ai;c1Sb�c1
ai;c2si� b�c2
� �¼ 1
Sb�c1
Xi
sibi;c1bi;c2 � Sb�c1b�c2
!
¼ 1Sb�c1
Xi
sibi;c1bi;c2 �Xi
sibi;c1b�c2�Xi
sib�c1bi;c2 þ
Xi
sib�c1b�c2
!
¼ 1Sb�c1
Xi
si bi;c1 � b�c1
� bi;c2 � b
�c2
� ¼Xi
si
S
oei;c1b�c1
oei;c2 ;
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230 229
where we have usedP
iai,c1=Sb̄c1 in the first equality; ai,c1= sibi,c1, ai,c2= sibi,c2, andPiai,c1=Sb̄c1 in the second equality; and Sb̄c1 =
Pisibi,c1, Sb̄c2 =
Pisibi,c2, and S =
Pisi in
the third equality.
References
Allen, F., Gale, D., 2000. Financial contagion. Journal of Political Economy 108, 1–33.
Berg, A., Pattillo, C., 1999. Are currency crises predictable? A test. IMF Staff Papers 46, 107–138.
Bernardo, A., Welch, I., 2004. Liquidity and financial market runs. Quarterly Journal of Economics 119,
135–158.
Borensztein, E., Gelos, G., 2003. A panic-prone pack? The behavior of emerging market mutual funds. IMF Staff
Papers 50, 43–63.
Broner, F., Lorenzoni, G., Schmukler, S., 2003. Why do emerging markets borrow short term? Mimeo, CREI.
Brown, K., Harlow, W., Starks, L.T., 1996. Of tournaments and temptations: an analysis of managerial incentives
in the mutual fund industry. Journal of Finance 51, 85–110.
Busse, J., 2001. Another look at mutual fund tournaments. Journal of Financial and Quantitative Analysis 36,
53–73.
Calvo, G., Mendoza, E., 2000. Rational contagion and the globalization of securities markets. Journal of
International Economics 51, 79–113.
Caramazza, F., Ricci, L., Salgado, R., 2000. Trade and financial contagion in currency crises. IMF Working
Paper 00/55.
Chen, H.L., Pennachi, G., 2002. Does Prior Performance Affect a Mutual Fund’s Choice of Risk? Theory and
Further Empirical Evidence. Mimeo, University of Illinois.
Chevalier, J., Ellison, G., 1997. Risk taking by mutual funds as a response to incentives. Journal of Political
Economy 105, 1167–1199.
Daniel, N.D., Wermers, R., 2000. Risk-Taking Behavior by Mutual Fund Managers: Do Managers dWalk AwayTfrom the Tournament? Mimeo.
Disyatat, P., Gelos, G., 2001. The asset allocation of emerging market mutual funds. IMF Working Paper 01/111.
Froot, K., O’Connell, P., Seasholes, M., 2001. The portfolio flows of international investors: I. Journal of
Financial Economics 59, 151–193.
Gelos, G., Wei, S., in press. Transparency and International Portfolio Holdings. Journal of Finance
Goldstein, M., 1998. The Asian Financial Crisis. Institute for International Economics, Washington DC.
Goldstein, I., Pauzner, A., 2001. Contagion of self-fulfilling financial crises due to diversification of investment
portfolios. Journal of Economic Theory 119, 151–183.
Grossman, S., Stiglitz, J., 1980. On the impossibility of informationally efficient markets. American Economic
Review 70, 393–408.
Harrison, J., Kreps, D., 1978. Speculative investor behavior in a stock market with heterogeneous expectations.
Quarterly Journal of Economics 92, 323–336.
Hernández, L., Valdés, R., 2001. What Drives Contagion: Trade, Neighbourhood, or Financial Links? IMF
Working Paper WP 01/29.
Hong, H., Stein, J., 2003. Differences of opinion, short sales constraints, and market crashes. Review of Financial
Studies 16, 487–525.
Ippolito, R., 1992. Consumer reaction to measures of poor quality: evidence from the mutual fund industry.
Journal of Law and Economics 35, 45–70.
Johnson, S., Boone, P., Breach, A., Friedman, E., 2000. Corporate governance in the asian financial crisis. Journal
of Financial Economics 58, 141–186.
Kaminsky, G., Reinhart, C., 2000. On crises, contagion, and confusion. Journal of International Economics 51,
145–168.
Kaminsky, G., Reinhart, C., Végh, C., 2003. The unholy trinity of financial contagion. Journal of Economic
Perspectives 17, 51–74.
Kaminsky, G., Lyons, R., Schmukler, S., 2004. Managers, investors, and crises: mutual fund strategies in
emerging markets. Journal of International Economics 64, 113–134.
-
F.A. Broner et al. / Journal of International Economics 69 (2006) 203–230230
Kodres, L., Pritsker, M., 2002. A rational expectations model of financial contagion. Journal of Finance 57,
769–799.
Koski, J., Pontiff, J., 1999. How are derivatives used? Evidence from the mutual fund industry. Journal of Finance
54, 791–816.
Kumar, M., Persaud, A., 2001. Pure contagion and investors’ shifting risk appetite: analytical issues and empirical
evidence. International Finance 5, 401–436.
Kyle, A., Xiong, W., 2001. Contagion as a wealth effect. Journal of Finance 56, 1401–1440.
Scharfstein, D., Stein, J., 1990. Herd behavior and investment. American Economic Review 80, 465–479.
Scheinkman, J., Xiong, W., 2003. Overconfidence and speculative bubbles. Journal of Political Economy 111,
1183–1219.
Schinasi, G., Smith, T., 2000. Portfolio diversification, leverage, and financial contagion. IMF Staff Papers 47,
159–176.
Shleifer, A., Vishny, R., 1997. The limits of arbitrage. Journal of Finance 52, 35–55.
Sirri, E.R., Tufano, P., 1998. Costly search and mutual fund flows. Journal of Finance 54, 359–375.
Van Rijckeghem, C., Weder, B., 2001. Sources of contagion: is it finance or trade? Journal of International
Economics 54, 293–308.
Van Rijckeghem, C., Weder, B., 2003. Spillovers through banking centers: a panel data analysis of bank flows.
Journal of International Money and Finance 22, 483–509.
World Bank, 2003. Global Development Finance. Washington, D.C.
Yuan, K., 2004. Asymmetric price movements and borrowing constraints: a rational expectations equilibrium
model of crises, contagion, and confusion. Journal of Finance 60, 379–411.
When in peril, retrench: Testing the portfolio channel of contagionA stylized modelDemandInelastic supplyElastic supplySome comments on the model
DataPortfolio dispersion over timeFund performance and portfolio choiceAn index of financial interdependenceFinancial interdependence and contagionConclusionsFund performance and redemptionsEquivalence of indices of interdependenceReferences